A Methodology Towards Integrated Urban Mobility

Solutions - Living PlanIT’s Case

Salvador Delgado Vaz Pinto

Department of Engineering and Management, Instituto Superior Técnico

Abstract

With the increasing rate of urbanization growth throughout the world, cities are faced with numerous challenges. Among them, urban mobility is a central one, and a key driver towards smart, sustainable cities, which in turn provide a higher quality of life for their citizens. The role of sustainable mobility, and its impact on society and the environment, is evident and recognized worldwide. Nevertheless, although there are a growing number of measures and projects that deal with sustainable mobility issues, there is a lack of tools capable of comparing their results and especially, of ensuring a holistic evaluation to facilitate replicability of the best practices. The methodology presented in this paper consists in the development of a model, based on the Multiple Criteria Decision Analysis discipline, capable of aiding in the decision-making of urban routes by users. This is performed by defining three fundamental components: 1) The set of alternatives from which the decision has to be made; 2) The set of criteria on which the alternatives are to be evaluated; and 3) The model itself, used to perform that evaluation. The output of this paper is a model that produces a ranking of routes referent to a specific urban journey. It encompasses ten distinct means of transport weighted against eight criteria. These criteria cover different angles of sustainable urban mobility and are applicable in different social and economic contexts, around the world. Furthermore, the model reflects the individual preferences of each distinct user, ensuring a personalized evaluation of urban sustainable routes.

Keywords: sustainable urban mobility, multiple criteria decision making, route optimization.

1. Introduction

The whole world’s population is increasingly concentrating in urban areas and this trend shows no sign of stopping. City policy makers have been facing several issues that arise from such a high population density and, among these issues, the increased demand for transportation of people and goods within cities is of key importance (UITP, 2015). The standard twentieth-first century city has plenty of mobility issues related with the movement of people and freight, beyond pollution. These issues also propagate their negative effects towards other important aspects of city life, thus degrading the overall quality of life of city dwellers (Navarro, 2016). It is t h e n

imperative to formulate an approach that takes into account the relationships between urban mobility and the daily city life. Mobility is a key part of everybody’s daily life and its needs are growing and changing all over the world. Individual mobility is increasingly reaching its limits due to the non-stop urbanization that spikes up the demand for urban mobility systems (Harris and Tapsas, 2006; Schrank et al., 2012). Mobility is also crucial for the good functioning of any city where its infrastructure is perceived as the number one priority area for cities to attract investors and also requires the biggest investment share among any other city sector (UITP, 2015). Cities rely on their transportation systems to manage both people and goods

1

2

between their multiple origins and destinations (Van Audenhove et al., 2014). Unfortunately, cities tend to struggle when faced with the pressures of modern mobility needs due to their outdated infrastructural systems and lagging transportation policies (Rodrigue, 2013). This makes them unable to cope with the high levels of concentration of economic and leisure activities that characterize cities and, in turn, results in several nefarious effects (Cohen and Muñoz, 2015). All of these factors lead to nefarious effects, such as extreme traffic congestion, parking issues, environmental impacts, energy waste, accidents and lack of safety, inefficient public transports, longer commuting and lack of comfortable mobility solutions (Rodrigue, 2013). With such a complex and broad set of issues concerning urban mobility, it is essential to properly establish a framework capable of methodically tackling said issues. To achieve this, key urban mobility dimensions are created, which facilitate the approach towards the problem. These

dimensions are summarized in Table 1,

alongside their respective issues for consideration.

Table 1 - Framework of urban mobility dimensions

and respective issues to consider

Dimension Issues to consider

Traffic Congestion

Energy and time wasted Aggravates other correlated issues (such as pollution, commuting, etc.)

Parking Issues Energy and time wasted Aggravates traffic congestion

Travel Times

Energy and time wasted Creation of disparity between professional opportunities

Comfort The physical and mental comfort that users experience

Safety and Security

The public health and security of city dwellers

Energy Consumption

Poor management of a resource already under extreme duress

Emissions

Pollution generated which aggravates climate change and health of citizens

Cost Affordability

This framework encompasses all of the issues targeted in this paper.

2. Living PlanIT

It is under this context that the problem under consideration arises, where Living PlanIT, a technology company that develops and implements through partnerships, technological solutions capable of providing a framework to establish Smart Cities. Specifically, the company wants to provide a model capable of improving urban mobility efficiency, whilst at the same time exploiting benefits derived from an integrated city wide approach. Living PlanIT, since its birth, has strongly pushed for the development of comprehensive, flexible, intelligent and scalable infrastructural solutions. Their most advanced output, the Living PlanIT Urban Operating SystemTM, is a revolutionary award-winning software platform developed to converge and manage the Internet of Things, that allows to manage city data and services. Living PlanIT has now been, for last seven years, on the forefront of the Smart City scene and has established partnerships with major companies and several governments in a wide sector dimension. A company with such a broad scope could not leave unattended the issues regarding urban mobility, on their quest of improving the world by making cities smarter.

2.1. Living PlanIT UOS

The PlanIT UOSTM is the revolutionary and award-winning software platform that Living PlanIT developed over the last years which has been designed to manage and converge the IoT in urban infrastructure. The UOS is a highly flexible, scalable, comprehensive, intelligent and all-encompassing middleware that delivers real-time sensing control, spatial analytics, support, security, data integration and provides a universal contextual framework required for every application. In simple terms, the UOS manages the city data and services in real-time. This is done by collecting data from sensors embedded in cities’ infrastructure and buildings, mobile devices and city hardware, then processing the gathered data streams and finally acting upon the resulting analysis. These resulting actions are as many as the imagination of the human mind allows (with the proper infrastructure and digital applications implementation), and can go from simple temperature adjustments in a room to the whole management of a city water and energy supply. Furthermore, the data and information collected is used to continuously improve the infrastructure’s performance and the UOS’ features. In technical terms, the UOS consists of a bunch of algorithms (the mathematical models that deal with the data) and software segments (that are responsible for

3

fetching information and processing it). The UOS analytics works in a Cisco medium and other similar hardware, such as routers and physical servers, where potential collaborators can jump in and develop new digital services. Since the UOS is a platform that acts as a “city brain” it is built and run on a rich set of open industry standards application programming interfaces (APIs), provides a configurable data model and entry points for custom logic. This is a core and essential feature of the UOS, since no city software platform should be locked into proprietary standards. With this flexibility and accessibility, the infrastructure owners are able to modify and personalize the platform for their specific application and hardware requirements. It also enables the development of applications by third party developers who can capitalize on the insights generated by the core layer of the UOS.

2.1.1. Living PlanIT UOSTM Architecture

The UOS architecture has a layered structure which can cope with an unlimited number of devices, users and sensors, regardless of their type. The global infrastructure environment where the UOS runs, as seen in Annex E, is structured in four layers: sensor/actuator network layer, controls layer, supervisory or core layer, and applications layer. Both the supervisory and control layers are the

The model, or method, to be used to effect the analysis.

The set of alternatives or options from which a choice (decision) has to be made;

The set of criteria against which the alternatives are to be evaluated;

3.1. Model Definition

To achieve the proposed goal, it is fundamental to identify a MCDA technique capable of tackling the problem. Out of the many techniques available, the Additive Model arises as the most adequate, given its straightforward intuitive appeal and transparency. Models of this type have a well-established track record of providing robust and effective support to decision-makers working on a range of problems and in various scenarios Roy (1996). The Additive Model establishes how the performance of an alternative on the many established criteria is merged into one global value. This is executed by multiplying the utility value of each criterion by the scale coefficient of that criterion, and then adding all those weighted scores together. The Additive Model is formulated as follows:

𝑽(𝒂𝒊) = ∑ 𝒌𝒊

𝒏

𝒊=𝟏

. 𝒗𝒊(𝒂𝒊)

𝒘𝒊𝒕𝒉 ∑ 𝒌𝒊

𝒏

𝒊=𝟏

= 𝟏 𝒂𝒏𝒅 𝒌𝒊 > 𝟎 (𝒊 = 𝟏, … , 𝒏)#1

true brain of the PlanIt UOSTM, described below: where:

Control layer: Contains all of the network devices implemented throughout the physical infrastructure, and manages the control codes and the driver applications responsible for bridging the diversified aggregation of sensors, users and devices. It also has a key role in administering the required immediate responses, such as the shutting down of an A/C unit once a certain temperature is reached.

Supervisory layer: Collects user and device data from the control layer, by managing and fusing it in sophisticated ways, and finally redistributes it across the ecosystem making sure it reaches its target sensor or device. The supervisory layer also boasts and API interface where infrastructure providers or third party developers can design and integrate applications, which are both location and context aware

3. Model Development

Roy (1996) defines the starting point for multiple criteria analysis as a well framed problem in which the three following components are clearly stated:

𝑉(𝑎𝑖) – Global value of alternative 𝑎𝑖

𝑎𝑖 – Performance of alternative 𝑎𝑖 in criterion i

𝑣𝑖(𝑎𝑖) – Utility value of the performance

𝑎𝑖 in criterion i

𝑘𝒊 – Scale coefficient of criterion i n – Number of criteria

3.2. Alternatives

The second logical step, after establishing the model utilized, is to characterize the available alternatives. Since the problem consists in optimizing a sustainable urban trip, it is necessary to consider the means of transport that city dwellers utilize. The following relevant means of transport are identified:

Private Car;

Motorbike;

Walk;

Bicycle;

Train;

Metro;

Tram;

Bus;

Ferry;

On-demand Ride Services;

4

With the complexity inherent of urban mobility ecosystems, it is fully expected that the options, or alternatives, present in the model won’t consist of only a single means of transport. Multimodal urban trips provide near-optimal solutions to city dwellers’ mobility needs. These heterogeneous alternatives take into account the individual characteristics of all of the means of transport that constitute said alternative.

3.3. Criteria

In order to perform the proposed multiple criteria analysis, it is of utmost importance to build a criteria tree that accurately represents the problem (Belton and Stewart, 2010). A criteria tree is like an objectives’ hierarchy, it shows how higher order objectives are linked to sub- objectives and eventually to performance measures or criteria. The construction of a criteria tree is a common practice once criteria have been properly identified. It helps to visualize the global approach towards the problem and to re-evaluate the criteria choices. For the problem under study, the urban mobility dimensions previously defined in section 1 serve as the foundation to build the criteria tree. These have been grouped into three main components: 1) Social impacts; 2) Environmental impacts; and 3) Financial impacts. The resulting work can be seen in Figure 1. These criteria are the measures of performance through which the options are evaluated by. One of the most important parts in adding value to a MCDA is the formal process of establishing a rationally based set of criteria, against which to judge the options (Belton and Stewart, 2010). Since the criteria act as the performance measures for the MCDA, they need to be operational. A judgement or a measurement needs to specify how well each option meets the objectives expressed by the criteria. It is fundamental to make sure that it is possible in practice to measure or judge how well an option performs, on the identified criteria.

The development of consistent numerical scales for the assessment of criteria is also a critical point towards the MCDA process (Bottero et al., 2017). Since the criteria in typical MCDA models are measured in diverse units, it becomes decisive to constitute normalized scales. This procedure is now described, for both performance measurements and scale coefficients.

3.3.1. Performance Measurements Bottero et al. (2017) introduce a modified version of Simos’ deck of cards method (Maystre et al., 1994) for calculating the utility value interval scales (usually within the range [0;1]) of criteria within the context of outranking methods. In the conclusion of their work, Bottero et al. (2017) state that the method can be adapted to build other ratio scales as well as to determine the criteria scale coefficients. The translation from the original criteria scales to a single common interval scale requires the use of a procedure that accounts for the intensity of preferences, between consecutive intervals of the scale. The objective of the current and the next subsection is to demonstrate how the author’s extension of the deck of cards method achieves this, for both interval and coefficient scales, respectively. In any case, the dialog between the analyst and the decision-maker is fundamental and, for utility value interval scales, should proceed as follows:

1. The analyst should provide the decision- maker with a discrete scale of criterion g: Eg = {l1,

l2,...,lk,…,lt}, where l1 ≺ l2 ≺ · · · ≺ lk ≺ · · · ≺ lt−1 ≺ lt (≺ means “strictly less preferred than”).

2. The analyst should also provide the decision- maker with a large enough set of blank cards.

3. The analyst should ask the decision-maker to define two reference levels, say lp and lq, and assign two utility values to these reference levels. These are frequently scored as u(lp) = 0 and u(lq) = 1. Alternative values can be assigned to lp and lq. Notice that very often levels lp and lq

coincide with l1 and lt, respectively.

4. The analyst alerts the decision-maker to the fact that two consecutive positions in the scale may be more or less close, in terms of his own preferences. This greater or smaller closeness is modelled via the insertion of blank cards in the intervals of the consecutive positions.

5. Consider the ranking of the levels with a certain number of blank cards, ek, in the intervals between every two consecutive levels, lk and lk+1, k = 1, . . ., t − 1:

l1 e1 l2 e2 · · · · · · lp ep lp+1 ep+1 · · · · · · lk ek lk+1 · · · · · · lq−1 eq−1 lq · ·· · · · lt−1 et−1 lt.

6. Now, consider only the levels in between lp and lq (in-between levels lk and lk+1 there are (ek + 1)

Figure 1 - Criteria tree concerning an Urban Sustainable Route

Environmental

Impacts

Parking Issues

Comfort

Financial Impacts

Cost

Travel Times

Sustainable Route

Social Impacts

Traffic Congestion

Safety and Security

Energy Consumption

Emissions

5

𝑗

𝑗

units) and compute the unit valuation: 𝑢(𝑙𝑞) − 𝑢(𝑙𝑝)

𝛼 = ℎ

where 𝑞−1

ℎ = ∑(𝑒𝑘 + 1) 𝑘=𝑝

utilities on each criterion are 0 and 1, respectively, they propose to build a hypothetical

2 reference set of alternatives. This set of alternatives (as many as the number of criteria) is constituted by alternatives that score 0 in all of

3 the criteria except for a single one, which is rated at its maximum utility value, 1. For each

which represents the number of units between levels lp and lq.

7. Finally, with the value of α, we can compute the utility value u(lk) for each level, as follows:

ℎ = ∑(𝑒𝑘 + 1)

𝑞−1

𝑘=𝑝

4

This procedure, of straightforward application for discrete criteria scales, can also be applied for continuous scales (such as cost or time). In order to do so, the analyst and the decision-maker

must together first define a minimum (𝑔𝑙 ) and a

maximum (𝑔𝑢) performance of a given criteria g.

The next step is to discretize the continuous scale in appropriate intervals. Subsequently, the deck of cards method is applied to this auxiliary discrete scale. Finally, to compute the proper utility value of the continuous scale a linear interpolation is used, as follows:

𝑢(𝑙𝑘) = 𝑢(𝑙1) + 𝛼(∑ (𝑒𝑗 + 1)𝑘−1

𝑗=1 ) , 𝑓or 𝑘 =2, , 𝑡. 5

3.3.2. Scale Coefficients Assessment

In order to compute a final combined value, it is necessary to establish the relationship (or scale coefficients) between the utility values of each criterion. The purpose of the criteria scale coefficients is to enable all scores to be converted to a common scale while accurately reflecting their relative importance to the decision. To achieve this, another modified version of Simo’s deck of cards method developed by Bottero et al. (2017) is utilized. According to the authors, the method should also begin with a dialog between the analyst and the decision- maker, as follows:

1. The analyst should provide the decision- maker with a first set of cards, where each card contains the name of each criteria and some additional information (if necessary).

alternative, the criteria that is ranked at his maximum value changes, whereas all of the other criteria value is set at 0. This leads to a set of alternatives whose tally is equal to the number of the criteria where each of the criteria are ranked at its maximum value only once. The decision-maker should then be confronted with this hypothetical set of alternatives and be prompted to rank each alternative in terms of his own preference. The outcome of the alternatives’ ranking produces the required criteria ranking in parallel.

4. Consider the ranking of the criteria provided by the decision-maker and denoted by r1, . . ., rh, . . ., rv, (r1 representing the least preferred criteria, r2 containing the second least preferred criteria, and so on, until rv, containing the most preferred

criteria). The analyst should now call the attention of the decision-maker to the fact that two consecutive positions in the criteria ranking may be more or less close in terms of importance. This greater or smaller closeness can be modelled through the insertion of blank cards in the intervals of the consecutive positions, in the criteria ranking. Let eh denote the number of blank cards between the criteria rh and rh+1, h = 1, . . ., v − 1.

5. The analyst should then obtain from the decision-maker information regarding the ratio z. z represents the ratio between the value of the most appreciated and the value of the least appreciated criteria (i.e., how many times the most appreciated criteria is more important than the least appreciated one).

6. Assign a value to criterion r1, say w(r1) = l. Typically, w(r1) = 1.

7. Compute the value of each unit, α, which is obtained by dividing the difference between the values of the most preferred criterion (w(rv) =

𝑙.z) and the least preferred objects (w(r1) = 𝑙 ) as follows:

2. The analyst should also provide the decision- maker with a large enough set of blank cards.

3. The analyst must then request that the decision-maker forms a ranking of the cards (criteria) in the first set. Marichal and Roubens

where

𝑙(𝑧 − 1) 𝛼 = 6

𝑠

𝑣−1

𝑠 = ∑(𝑒ℎ + 1) 7 ℎ=1

(2000) propose a method to aid in this criteria ranking definition. Based on their idea, and assuming that the minimal and the maximal

which represents the number of units between r1

and rv.

6

∑

8. Compute the values of w(rh) for h = 2,…, v, as

follows: ℎ−1

𝑤(𝑟ℎ) = 𝑙 + 𝛼 (∑(𝑒ℎ + 1)) , for ℎ = 2, … , 𝑣. 8

𝑗=1

This results in the scale coefficients correspondent of each criterion.

9. Finally, to compute the normalized scale coefficients of each criterion, the following formula applies:

𝑤(𝑟ℎ )

For the sake of the analysis we’ll consider that the user has at his disposal a private car, a motorbike and a bicycle in the starting point A, Saldanha Square. By inputting Scenario A into the model developed, it is possible to compute the global performances of each route. These have been calculated through the Additive model described in sub-section 3.1. The final results, sorted by rank, follow in Table 3. According to the model, the most preferred route for the user is to ride a bicycle, take the bicycle with him in the red line metro, swap to

4. Results

𝑢𝑘 = 𝑣

𝑗=1 w(𝑟𝑗)

9 the green line metro and then ride again the bicycle towards the final destination. Both Cost and Travel Times criteria have the highest influence in boosting Route 7 to the top position, which makes sense since they’re they two

To validate the model developed, three relevant scenarios are analysed in the city of Lisbon. The proposed goal is to define a sustainable route between two given urban points. For the three scenarios, the feasible routes are established in terms of means of transport and geographical routing. These are submitted to the model so that they are evaluated and ranked. With the overall ranking of each route, the optimal sustainable route is established, considering the user preferences. The user preferences have been computed by applying the method described in sub-section 3.3.2. and results in

the scale coefficients presented in Table 2 :

Table 2 - Scale Coefficients of each criteria

Criteria Scale Coefficient

Cost 0,250

Travel Times 0,219

Comfort 0,168

Safety and Accidents 0,147

Traffic Congestion 0,106

Parking Issues 0,064

Energy Consumption 0,033

Emissions 0,013

4.1. Scenario A

The first scenario is a common inner city route in Lisbon. The user is in point A, the Saldanha square, and wishes to go to point B, the Dom Luís Garden in Cais do Sodré. For this specific trip, the user can resort to several different routes with distinct means of transport, specified in Annex A. It is of key importance to define which options, in terms of both means of transport and geographical routes, the user can survey.

criteria with the highest scale coefficient. Inversely, the Comfort criterion, which features the third highest scale coefficient, barely contributes to the global performance of the route. This is justified by its low performance which is corresponded with a 0,222 utility value.

4.2. Scenario B

The second scenario under analysis is a common commute from home to work for many Lisbon city dwellers. The route begins in the Lisbon South Bay in Almada. Specifically, point A is Rua Liberdade, 67. The final destination, point B, is in the heart of Lisbon, the Luís de Camões Square. For this scenario the user has, yet again, plenty of distinct routes and different means of transportation at his disposal. Since it is a longer journey, compared to scenario A, the broader the set of options the user has to choose from. Concurringly to what happened in Scenario B, we admit that the user has a car, a motorbike and a bicycle in the starting point A, Rua Liberdade 67. With all of the feasible routes established, specified in Annex B, Scenario B is inputted into the model. By computing the utility values of all of the criteria for each route under analysis, the global scores of Scenario B, derived from the model, can be seen in Table 4, sorted by rank. For this scenario, Route 4 ranks the highest which consists of riding a bicycle from point A, taking the ferry with the bicycle and then riding the bicycle until the Luís de Camões Square. This route ranks so highly due to its low cost (1,98€) and short journey duration, with an estimated duration of 21 minutes.

Table 3 - Global performance of Scenario A

Route R7 R3 R6 R4 R1 R2 R10 R8 R5 R9

Global Score

0,771 0,767 0,757 0,707 0,688 0,639 0,634 0,629 0,624 0,624

7

Table 4 - Global performance of Scenario B

Route R4 R6 R5 R14 R11 R7 R10 R15 R16 R13 R18

Global Score 0,702 0,684 0,678 0,671 0,670 0,662 0,658 0,657 0,642 0,640 0,635

R12 R8 R1 R9 R17 R3 R19 R23 R20 R21 R2 R22

0,631 0,631 0,617 0,600 0,599 0,563 0,544 0,544 0,544 0,537 0,528 0,487

4.3. Scenario C

The third and final scenario is a common touristic and leisure journey. The user is in point A, the Pastéis de Belém Store in Rua Belém 84 and wishes to go to Lx Factory in Rua Rodrigues de Faria 103 which is point B. For this trip, both the bus and tram stop are in such a short distance that the user only considers walking to reach it. If he wishes to use a different means of transport (such as a private car or bicycle) he will complete the whole trip with it, since the destination point B is very near to him as well (2,8km). As performed previously, it is assumed that the user has a private car, a motorbike and a bicycle at his disposal in point B. With the feasible routes established, presented in Annex C, the model evaluates each one of them. By computing the utility values of all of the criteria for each route under analysis, the global score of each route is determined and presented in Table 5, sorted by rank. For this scenario, the Bicycle ranks the highest out of all the options. This is due to the inexistent cost of this journey combined with a relatively short trip duration (10 minutes).

4.4. Sensitivity Analysis After the analysis of the three scenarios and their respective results, a crucial part of the MCDA process unfolds, the sensitivity analysis. This can be performed in two distinct ways: 1) By altering the way options’ performances and utility values are measured; and 2) By adjusting the scale coefficients, in order to test the same options with different user preferences. Since the performances’ measurements of each means of transport and the respective conversion into utility values are inherent to the means of transport themselves, it has been established that the focus of this sensitivity analysis are the scale coefficients. This decision is further backed by the intention of developing a model capable of being personalized, hence the vantage point of analysing its behaviour with alternative user preferences.

The three exact same scenarios, A, B and C, are the foundation of this sensitivity analysis. These are analysed under the preferences of two new distinct users: 1) A user that has no preferences whatsoever. For him, every criterion has the exact same importance when planning an urban journey. This user is henceforth addressed as user I; and 2) A user with a substantial environmentalist preoccupation and less focused on personal gratification criteria such as Cost, Travel Times, Parking Issues and Comfort. This user is henceforth addressed as user E. By repeating the process described in sub-section 3.3.2, the new users respective scale coefficients are defined, as visualized in Table 6. An analysis on how each scenario reacts, when faced with the new scale coefficients inputted in the model, is performed.

Table 6 – User I and User E respective scale coefficients

Criteria

User I

User E

Emissions

0,125

0,233

Energy Consumption

0,125

0,208

Safety and Accidents

0,125

0,171

Traffic Congestion

0,125

0,147

Comfort

0,125

0,097

Parking Issues

0,125

0,073

Travel Times

0,125

0,048

Cost

0,125

0,023

After analysing the model output, with the new user preferences, some interesting changes arise in Scenario A, as seen in Table 7 and 8. For both the news users Route 6 becomes the highest ranked one. This route consists on walking, taking the red line, the green line metro, and then walking to the end destination. It’s a very similar route compared to Route 7 (ranked in first place for the original user), which maintains a strong ranking for both news users, where the walking stretches are replace by riding a bicycle. Although being ranked as the top route for both users, Route 6 accomplishes this performance due to distinct traits in the eyes of the two new users.

For user I, this route is considered optimal due to a strong performance in the following criteria: 1) Parking Issues, where it has a maximum

Table 5 - Global performance of Scenario C Route R4 R3 R5 R1 R2 R6 R7

Global Performance 0,776 0,746 0,732 0,703 0,652 0,647 0,614

8

performance of 1 that corresponds to a utility value, of 0,125; 2) Traffic Congestion, with a near maximum utility value of 0,121; 3) Safety and Security, with a utility value of 0,120; 4) Emissions, with utility value of 0,116; 4) Energy Consumption, with a utility value of 0,112 corresponding to an energy consumption of 0,0329 koe; and 5) Cost, with a utility value of 0,109 that corresponds to a cost of 1,45€.

On the other hand, for user E, the route peaks his route preferences through the robust performance of these criteria: 1) Emissions, with a utility value of 0,216; 2) Energy Consumption, with a utility value of 0,187 (0,0329 koe); and 3) Safety and Security, with a utility value of 0,164. Even with fundamental variations on the scale coefficients, Routes 3 (Bicycle), 6 (Walk + Metro + Metro + Walk) and 7 (Bicycle + Metro + Metro + Bicycle) manage to rank always in the top three positions, for all of the users. This is a great indicator that these three routes are near-optimal for Scenario A.

Scenario B, with its complex routes’ configuration, varies considerably after inputting the new two users’ data, as seen in Table 9 and 10. Route 14 (Bicycle + Train + Bicycle + Metro + Bicycle), by escalating three positions, becomes the top ranked route for the two users. As in scenario A, the motives vary between users. For user I, Route 14 ranks as number one due to strong performances in the criteria Energy Consumption, Traffic congestion, Emissions, Safety and Security and Parking Issues. Alternatively, user E perceives Route 14 as his best choice due to strong performances in terms of: 1) Emissions; 2) Energy Consumption; 3) Safety Security and; 4) Traffic Congestion.

Table 7 - Global performance of Scenario A for user I

Route R6 R3 R7 R8 R9 R10 R5 R4 R2 R1

Global Score 0,819 0,817 0,812 0,673 0,664 0,655 0,642 0,609 0,598 0,499

Variation +2 0 -2 +4 +5 +1 +2 -4 -3 -5

Table 8 - Global performance of Scenario A for user E

Route R6 R7 R3 R8 R9 R10 R5 R2 R4 R1

Global Score 0,859 0,851 0,829 0,691 0,675 0,665 0,636 0,502 0,500 0,417

Variation +2 -1 -1 +4 +5 +1 +2 -2 -5 -5

Table 9 - Global performance of Scenario B for user I

Route R14 R11 R15 R10 R16 R18 R13 R12 R4 R7 R5

Global Score 0,770 0,768 0,762 0,761 0,726 0,724 0,724 0,718 0,717 0,652 0,642

Variation +3 +3 +5 +3 +4 +5 +3 +4 -8 -4 -8

R6 R8 R17 R9 R19 R23 R20 R21 R3 R2 R22 R1

0,635 0,630 0,612 0,610 0,606 0,594 0,592 0,579 0,533 0,477 0,473 0,460

-10 0 +2 0 +2 +2 +2 +2 -3 +1 +1 -9

Table 10 - Global performance of Scenario B for user E

Route R14 R11 R15 R10 R16 R18 R12 R13 R4 R17 R9

Global Score 0,855 0,852 0,852 0,850 0,791 0,784 0,783 0,780 0,715 0,712 0,707

Variation +3 +3 +5 +3 +4 +5 +5 +2 -8 +6 +4

R19 R20 R23 R7 R21 R5 R8 R6 R22 R3 R2 R1

0,638 0,628 0,617 0,615 0,609 0,599 0,594 0,567 0,544 0,480 0,458 0,404

+6 +7 +5 -9 +5 -14 -5 -17 +3 -4 0 -9

9

Scenario C, a common touristic and leisure journey between the Pastéis de Belém shop and Lx Factory, is now put to test under the new preference conditions of the two novel users. The performance of Scenario C, under this revamped model, is presented in Table 9 and 10.

Although the scale coefficients defined by user I and user E vary considerably, both users obtain a similar set of route rankings, where only Route 2 and 5 do not coincide, as seen in Tables 11 and 12. Route 3, which consists on only walking, improves its ranking by one position and stands out as the highest performing route for in both cases. User I perceives this route as optimal due to the strong influence of the following criteria: 1) Parking Issues, with a maximum utility value of 0,125; 2) Energy Consumption, as well with a maximum utility value of 0,125; 3) Cost, which also features a maximum utility value of 0,125, since the cost of the route is null; and 4) Emissions, with a near maximum utility value of 0,123. At the same time, user E values Route 3 as optimal since it performs solidly in the Emissions and Energy Consumption criteria

With these new preferences set, the model output demonstrates a clear trend, for all the scenarios: routes that feature greener means of transport, such as walking, riding a bicycle and all of the public transportation, experience considerable increases in their ranking. This translates in the selection of more sustainable routes by the two new users. Furthermore, and the most important conclusion of this section, is the consistently dominance of the Bicycle over other means of transport. The importance of bicycles in the development of sustainable urban mobility is praised by several urban mobility publications, and the model evidently confirms this trend. The strong bicycle performances result in being featured in the first position in five out of the nine analysis performed (three scenarios with three distinct users). In the remaining four analyses, it is featured in the second position at a mere distance of 0,02 and 0,08 in scenario B, for user I and E, respectively. In scenario C, the distance to the first position is valued at 0,02 and 0,014 for user I and E, respectively.

In low distance routes (A and B), on-demand ride services outrank the private car. The reverse is experienced when analysing longer routes, such as scenario B. This is applicable regardless of the user’s preferences being utilized.

5. Conclusions Recent exponential growth in urbanization rates has generated numerous challenges for both city policymakers and city dwellers. Among them, the imbalance between urban mobility supply and demand, of both people and goods, has become a pressing issue, since mobility is crucial for the good functioning of cities and their respective citizens’ quality of life. In this paper, eight fundamental urban mobility dimensions are identified: 1) Traffic Congestion; 2) Parking Issues; 3) Travel Times; 4) Comfort; 5) Safety and Security; 6) Emissions; 7) Energy Consumption; and 8) Cost, to better comprehend and consequently framework the problem under study. The model development is the central part of this paper. The process respects the traditional stages of the MCDA discipline, described in section 3. The primary alternatives of the model are the following means of transport: 1) Private Car; 2) Motorbike; 3) Walk; 4) Bicycle; 5) Train; 6) Metro; 7) Tram; 8) Bus; 9) Ferry; and 10) On- demand Ride Services. The actual alternatives end up being usually a combination of these ten elements, due to the synergies present in multimodal trips. The criteria of the model reflect the urban mobility dimensions defined in section 1. The selected performance measurements, for all of the criteria, are the core of the evaluation methodology towards urban sustainable mobility. For each of the criteria, data is gathered in order to calculate the performance value of each alternative. This process is aided by existing databases and research for criteria with objective quantitative date, while for the criteria with a subjective component the dialog between the analyst and the user is essential. The third component is resolved through the utilization of the Additive Model.

Table 11 - Global performance of Scenario C for user I

Route R3 R4 R7 R6 R5 R2 R1

Global Score 0,810 0,808 0,693 0,663 0,623 0,606 0,508

Variation +1 -1 +4 +2 -2 -1 -3

Table 12 - Global performance of Scenario C for user E Route R3 R4 R7 R6 R2 R5 R1

Global Score 0,836 0,822 0,740 0,656 0,504 0,504 0,420

Variation +1 -1 +4 +2 0 -3 -3

10

The analysis of three distinct scenarios is detailed in section 4. This enables the corroboration of the successful functioning of the model and the drawing of conclusions regarding urban journeys analysed. Three distinct scenarios are evaluated through the preferences of three different users, each with alternative views on what a sustainable route is. This enables judging the model’s performance when facing clashing preferences, and better understanding what factors truly matter in an urban sustainable route. The main conclusions drawn from this process are the following: 1) Bicycles are an outstanding urban means of transport and should be encouraged. They perform consistently with high ranks, for different scenarios and different users. 2) If the user has little regard for the Environmental Impacts group of criteria (Emissions and Energy Consumption) public transport, although feasible in some cases, is rarely ideal. By changing user preferences to, at the very least, rate the Environmental Impacts’ criteria as equal as the other criteria, the public transport is immensely boosted and is rated frequently as an ideal solution. 3) Irrespective of the user preferences being used, the strength of multimodal trips is always present. These enable synergies between distinct means of transport which benefit the user tremendously. As a main conclusion, the reliable performance of the model is praised. The model functions accordingly to its creation purpose: It outputs, based on user preferences, a ranking of routes based on their perceived sustainability, in a given journey from A to B. Although this is the prime goal of the model, it can further be utilized, with some minor tweaks, for other relevant purposes, such as: 1) Self-evaluation of a city’s mobility system: what is the global score, and what are the strengths and weaknesses? This results in a specific city profile; 2) Monitoring: a regular update of the data present in the model shows the evolution of aspects concerning the sustainability of the urban mobility system, portraying the impact of measures taken; 3) Policy assessment: by estimating the impact of a planned (package of) mobility measure(s) on each of the criteria, insight into the global sustainability of the proposed mobility investments is retrieved; and 4) Benchmarking with other cities: evaluating how the performance measurements and scale coefficients compare to other cities. This allows the city to set (realistic) targets for its future development; The flexibility of the model is also noteworthy. Since it utilizes the Additive Model, the incorporation of say, new criteria, is facilitated

by its inherent simplicity. This also applies to the case of adding new means of transport. 6. Future Research. The implementation of the model developed into Living PlanIT’s UOS is a promising merger. Backed by the UOS real-time sensing of data, the model can improve its gathering data process, thus providing more accurate route rankings. This integration requires the development of a strategy on two fronts: 1) The IoT deployment; and 2) The software solutions. The data collected by the UOS, through the utilization of the model, serves to map users’ urban travelling patterns. This data, properly processed, enables a consolidated prediction on the actual and future urban mobility needs of a city. This partnership also enables the possibility for the model to self-improve, based on machine learning. Through the regular use of a city dweller, the model can record the historical data and better adjust itself for future route requests. The model can also request user feedback in order to accelerate the learning process. With a perfection of the model, it can also be included in common route mapping apps, like Google Maps and Waze. Likewise, it can be utilized by public transportation companies, like Carris, in order to better promote their services. The model does not reflect the complete reality in several parameters due to the complexity of the problem itself, in the real world. One of the key features that the model lacks is the ability to compute the route’s rankings considering the interaction between criteria. Logically, a route that has a higher performance in the Travel Times criterion is prone to have a higher performance in criteria such as Energy Consumption or Costs. To model this, the Cochet multiple criteria preference model, developed by Bottero et al., (2017), is an interesting prospect to apply in the model developed. The Cochet integral enables the calculation of Route rankings by taking into account the interaction experienced between the criteria. This interaction can be positive, null (as is assumed in the model developed) and negative. As a final conclusion, it is expected that the resulting model motivates the further development of tools with the aim of improving urban mobility. Furthermore, with the implementation of the model in the UOS, it is also expected that synergies between distinct urban sectors and urban mobility arise, due to the crossover and processing of inter-sector data. This a fundamental factor in the transformation of cities into Smart Cities, that is yet to be explored.

11

7. References Belton, V., Stewart, T., & Belton, V. (2014).

Problem Structuring and Multiple Criteria Decision Analysis Problem Structuring and MCDA. http://doi.org/10.1007/978-1-

4419-5904-1 Bottero, M., Feretti, V., Figueira, J.R. & Roy,

B. (n.d.). Cahier du Lamsade 378 - On the Choquet multiple criteria preference aggregation model: Theoretical and problem insights from a real-world application

Cohen, B., & Muñoz, P. (2015). Sharing cities and sustainable consumption and production: towards an integrated framework. Journal of Cleaner Production, 97, 1–11. http://doi.org/10.1016/j.jclepro.2015.07.13 3

Harris, A., Tapsas, D., (2006). Transport and Mobility: Challenges, innovations and improvements. Royal Automobile Club of Victoria (RACV) Ltd.

Navarro, C., Furió, S., & Estrada, M. (2016). Designing new models for energy

efficiency in urban freight transport for smart cities and its application to the Spanish case. Transportation Research Procedia, 12, 314–324. Rodrigue, J. P. (2013). The Geography of

Transport Systems (3rd Edition). New York: Routledge.

Roy B. Multicriteria Methodology for Decision Aiding. Kluwer Academic Publishers, Dordrecht, 1996.

Schrank, D., Eisele, B., & Lomax, T. (2012). TTI’s 2012 Urban Mobility Report. Texas A&M Transportation Institute.

UITP, 2015 - Public Transport Trends, http://www.uitp.org/sites/default/files/cck- focus-papers- files/UITP_Trends_Exec_summary_12p. pdf, accessed on 26/1/2016

Van Audenhove, F., Korniichuck, O., Dauby, L., Pourbaix, J. (2014) The Future of Urban Mobility 2.0: Imperatives to shape extended mobility ecosystems of tomorrow. Arthur D Little in association with UITP.

12

8. Annexes

ANNEX A - Options schematics of Scenario A

Table 134 - Options schematics of Scenario A

Route Means of Transport

R1 Private Car

R2 Motorbike

R3 Bicycle

R4 On-demand Ride Services

R5 Walk + Bus + Walk

R6 Walk + Metro + Walk

R7 Bycicle + Metro + Bycicle

R8 Walk + Metro + Bus + Walk

R9 Walk + Bus + Walk

R10 Bicycle + Bus + Walk

13

ANNEX B - Options schematics of Scenario B Table 14 - Routes' schematics of Scenario B

Route Means of Transport

R1 Private Car

R2 Motorbike

R3 On-demand Ride Services

R4 Bicycle + Ferry + Bicycle

R5 Private Car + Ferry + Walk

R6 On-demand Ride Services + Ferry + On-demand Ride Services

R7 On-demand Ride Services + Ferry + Walk

R8 Motorbike + Ferry + Walk

R9 Private Car + Train + Train + Walk

R10 Bicycle + Train + Train + Walk

R11 Bicycle + Train + Train + Bicycle

R12 Motorbike + Train + Train + Walk

R13 On-demand Ride Services + Train + Train + Walk

R14 Bicycle + Train + Bicycle + Metro + Bicycle

R15 Bicycle + Train + Walk + Metro + Walk

R16 Motorbike + Train + Walk + Metro + Walk

R17 Private Car + Train + Walk + Metro + Walk

R18 On-demand Ride Services + Train + Walk + Metro + Walk

R19 Walk + Bus + Bus

R20 Bicycle + Bus + Bus

R21 Motorbike + Bus + Bus

R22 Car + Bus + Bus

R23 On-demand Ride Services + Bus + Bus

14

ANNEX C - Options schematics of Scenario C

Table 14 - Routes' schematics of Scenario B

Route Means of Transport

R1 Private Car

R2 Motorbike

R3 Walk

R4 Bicycle

R5 On-demand Ride Services

R6 Walk + Bus + Walk

R7 Walk + Tram + Walk